https://ir.unisa.ac.za/handle/10500/2916 2026-05-14T07:55:09Z 2026-05-14T07:55:09Z Netshithuthuni, Rinae https://ir.unisa.ac.za/handle/10500/31951 2024-11-28T06:16:31Z 2024-02-23T00:00:00Z

Long-term stability analysis of open pit mine slopes : a case study of Wearne mines in Limpopo Netshithuthuni, Rinae The main objective of the study was to conduct a long-term stability analysis of the slope, which due to blasting and continuous mining activity may deteriorate in its strength and since there is very little slope stability monitoring done, the analysis is warranted. Several methods were applied to achieve the objectives. These methods include Kinematic analysis, Limit equilibrium and Numerical modelling. The kinematic analysis method made use of the discontinuity orientation and the slope orientation to determine the probability of different types of failures occurring. These failures include planar, wedge and toppling. Limit equilibrium method was used to determine the Factor of safety (FoS) of the slope by analysing the driving forces and the resisting forces of the slope. OPTUM G2 which is a numerical method was then used to simulate the failure of the slope in the presence of different length or depths of discontinuities. Lastly the ROCFALL which is also a numerical model was also used to determine the extent at which the rock will travel down the slope in case of failure. The results produced by the kinematic analysis showed that the slopes are most likely to experience toppling failure more than planar and wedge failures. The Limit equilibrium results on the other hand showed that the slopes were stable based on its FoS value. OPTUM G2 proved that the increase in the length of discontinuities reduces the strength of the slope. Lastly, the ROCFALL models showed that the slopes geometries are well on balance and the presence of safety berms will prevent the rock rolling to the lower benches.

2024-02-23T00:00:00Z Egling, Theodore https://ir.unisa.ac.za/handle/10500/31766 2024-10-20T10:22:02Z 2024-08-29T00:00:00Z

Differential evolution optimization algorithms and its application in machine learning based disease detection Egling, Theodore Machine learning (ML) in healthcare is crucial for establishing and enhancing performant automated disease detection systems that can be used in medical practice. This dissertation aims to improve ML classifiers for disease detection through three main objectives: optimising hyperparameters using Differential Evolution (DE), comparing the performance of DE-optimised classifiers (Random Forest, AdaBoost, Gradient Boosting) against traditional methods, and exploring DE's application in automating triage systems in healthcare. Research employed datasets from the UC Irvine Machine Learning Repository, focusing on heart disease and thyroid cancer due to their prevalence. Data analysis involved optimising classifiers with DE and assessing their performance. The study identified the DE-optimised Random Forest classifier as particularly effective, achieving 98.7% accuracy and a 97.2% F1-score in thyroid cancer recurrence, and 93.3% accuracy with a 90.9% F1-score for heart disease. These results underscore DE's potential in enhancing accuracy and efficiency of ML classifiers. The findings also suggest significant implications for DE in healthcare, especially in automated triage systems, indicating a transformative impact on predictive diagnostics. The dissertation concludes with recommendations for integrating DE-optimised ML classifiers in healthcare settings and suggestions for future research. Text in English

2024-08-29T00:00:00Z Lafleni, Sipho Pelican https://ir.unisa.ac.za/handle/10500/31477 2024-09-06T10:40:54Z 2024-06-05T00:00:00Z

Intelligent fault detection technique for distribution network with interconnected distributed generation sources Lafleni, Sipho Pelican Machine Learning (ML) or Artificial Intelligence (AI) approaches in a distribution protection system are proposed to detect and categorize distribution network (DN) issues with Distribution Generation. This study highlights an important but generally overlooked distribution generation feature in energy transformation. Distributed generators (DGs) integrated into Distribution Network (DN) capability improve power quality, system dependability, voltage sags, and emergency backup during protracted grid outages. A technical and global analysis of DG technology's increased penetration is revealing its effects such as the utility systems undergoing elevated fault current and load flow changes, which will affect current protective relaying, particularly overcurrent relays. A protective system that can respond to the new dynamic DN is required to avoid consequences. Intelligent protection system application is a suitable distribution network solution for the explained challenge. The study conducts a thorough assessment of various methods used in detection and diagnostic systems inside distribution networks that have interconnected distributed generators (DGs). The study specifically emphasizes the implementation of intelligence-based approaches. This evaluation is conducted through a thorough literature review. A comprehensive model of a distribution network, encompassing all essential components, was constructed and subjected to simulation. This model incorporated all pertinent parameters associated with the distribution network. Subsequently, several forms of faults were intentionally introduced at a specific point inside the micro-grid. The purpose of this exercise is to gather voltage and current signals at the busbar and then the collected data is subsequently transformed into numerical values to facilitate machine-learning modelling. The implementation of intelligent approach for fault detection in distribution networks with various machine-learning techniques, allows the approach to form part of the objective to gather related signals that are pre-processed as variable features in order to extract required data that can help identify distribution network and classify faults at most efficient and accurate way. The primary objective of the protection system was to analyze the underlying failure mode, determine the fault type quickly and accurately, and identify the faulty line in the system. S shows the micro-grid's successful operation with current and voltage signals evidently shown . The current and voltage signals are transformed to numerical values for feature extraction which is key requirement for machine learning modelling. The derived variable features from feature extraction were trained and tested to validate fault diagnosis and classification to find the best machine learning fault classifier, Support vector machine (SVM) classifiers shows excellent results with 99.9% accuracy in validation and testing. These accuracy results meet the difficult requirements of a micro grid protection systems and SVM's ability to simulate non-linear decision boundaries, which is valuable in many applications. Text in English

2024-06-05T00:00:00Z Chauke, Tiyani https://ir.unisa.ac.za/handle/10500/31328 2024-06-26T09:01:47Z 2023-03-01T00:00:00Z

Implicit modelling of geometallurgical parameters for application in mine planning Chauke, Tiyani The objective of the thesis was two-fold: to investigate the potential of implicit modelling techniques for modelling geometallurgical parameters in mine planning and to generate synthetic geometallurgical data using Generative Adversarial Networks (GAN) models. Several geometallurgical parameters, including ore grade, Bond work index (BWI), rod mill index, rock quality designation (RQD), drop weight index (DWI), Axb, and Abrasion index (Ai), were modelled in this thesis using implicit and geostatistical methods, and their results were compared. To generate synthetic geometallurgical data, GAN-based models were used, namely, Conditional Tabular Generative Adversarial Network (CTGAN), Copula Generative Adversarial Network (CopulaGAN), and Gaussian Copula. The process was conducted in Python® using a Synthetic Data Vault (SDV) library, based on original geometallurgical data obtained from previous research papers, theses, and online databases. Geometallurgical block models were produced using implicit (Radial Basis Function) and geostatistical (Ordinary Kriging) methods and compared. The following software packages were used in this study. Leapfrog® Geo, Microsoft® Excel®, and Microsoft® Paint®. The results of the Geometallurgical Block Model (GMBM) were compared using parameters of mine planning, such as the grade-tonnage curve and resource estimations. In conclusion, the study found that the synthetic geometallurgical data generated in this research was of high quality and demonstrated that implicit modelling methods can improve the accuracy and efficiency of mine planning by modelling geometallurgical parameters. However, more research is recommended to explore other implicit-based methods, such as the potential field and the Hermite Radial Basis Function (HRBF).

2023-03-01T00:00:00Z